Hi, everybody, Tim Horan, the cloud and communications analyst, here at Oppenheimer. My pleasure to be hosting the CEO of Cogent, Dave Schaeffer. I must admit, this is a first for me, but I've seen Dave do a lot of first things. Cogent is reporting earnings tomorrow, and Dave has been kind enough not to cancel, coming to our conference here, virtually. We are in no way gonna talk about the quarter whatsoever, because we, we can ask Dave all those questions tomorrow morning.
I have recently, this week, I have a 25-year comp sheet I keep, and I have moved Dave out of the communications sector up into the cloud infrastructure sector of my comp sheet, thanks to his acquisition here of T-Mobile, where he's gonna be doing a lot more fiber interconnectivity from cloud to cloud, all of this enabling AI. I really was looking forward to this conversation because Dave knows more about cloud and what's gonna be required by it for AI infrastructure than anyone you're gonna talk to. I think, I personally think we're in to see a pretty big upgrade cycle on CapEx globally, and Dave is gonna tell us how his company, he's gonna profit from that, and maybe even some other ways to profit from it. Dave, I know there was.
There wasn't a question in there. Do you have to start out with any safe, harbor statements or anything?
No, no safe harbor, Tim. First of all, I wanna thank you for inviting me. I wanna thank investors for their time. You know, I think I've been coming to Tim's conference now for a couple of decades, and I would not miss it for anything. Tim's always very insightful in the questions he asks. The only reason we changed the cadence of our reporting this year is because of the very acquisition that Tim mentioned. We acquired the former Sprint Global Markets Group physical network and operating business from T-Mobile. As part of that transaction, we received $700 million in cash payments from T-Mobile for providing transit services.
It's highly unusual to have a company acquire an asset where there was over $20 billion of hard capital invested, and then pay $1 for it and have the seller then give them $700 million. We had some clarifications we needed to get from the Securities and Exchange Commission on how to report, and that was the reason for the delay. You know, we look forward to talking about these thematic topics, and I think Tim's right. We are at an inflection in our industry. You know, Cogent is an internet service provider first and foremost, and today, most data in the world flows over the public internet. There are certain types of data that use other technologies. They may actually be more expensive and less flexible, but better suited for low-latency, deterministic traffic flows, and AI fits into that grouping.
So Cogent.
I'm sorry, Dave, to cut you off. Can you describe that type of transport again? Sorry.
Yeah. It's deterministic, meaning you have a point A and a point B, and the traffic flows always through that exact path with an exact amount of time, latency, associated with that physical path. On the Internet, traffic is routed, so the Internet works by using transport networks with a series of routers distributed, and the router only asks one simple question of the packet: "What is the best next hop to get you to your final destination?" There is no concept of an end-to-end flow, but rather a series of disjoint links, and that was by design, in order to survive a nuclear attack or any kind of disruption. If you are trying to replicate huge amounts of data, the Internet is not the best for that. Maybe the cheapest, but it is not the best, and there are two reasons.
One, the data may get between those two points at different intervals. The latency may change. Secondly, the Internet, by design, supports relatively small flows of information, typically packet sizes of about 1,500 bits. Whereas if you're doing a huge data replication between two big data centers, you may want to put data into what are known as jumbos, packets that are up to 9,600 bits, and you want to have very large amounts of those packets flowing simultaneously, and that is not good for Internet. As a result, we are repurposing the original network that Sprint built to carry long-distance voice, if you remember what that was. It was the first fiber optic network. It was built at a time when all long distance was delivered over fixed microwave and Sprint was at the forefront. Over a 40-year period, that network has effectively become dormant.
When T-Mobile acquired it, they knew it was not strategic. They also knew the large enterprise customers that Sprint was serving were not strategic to them, and that is why they sold us that business, and it is why we can repurpose both the network and the 1.6 million square feet of telco central office space that we acquired into data centers. For anyone who's followed AI, they know how power and space-intense AI computations are.
How much data center did you get again?
We ended up acquiring 482 facilities, with 1.9 million sq ft. 45 of those facilities we have earmarked for data center conversion. That is 1.3 million out of the 1.9, and we are going to end up with about 160 megawatts of power. That 160 megawatts added to the 70 megawatts that Cogent already has in its data center footprint, and the 45 incremental facilities will give us a grand total globally of just under two million sq ft, 220 megawatts of power across that footprint that will sell for raw colo space and power, and which that could be used to support AI, and then tied to the transport network that we acquired for data replication.
Dave, if you don't mind me asking, I mean, won't that cost, like, $1 billion to upgrade?
It won't, Tim, and the reason is, these facilities were built to a very high standard. They all have redundant power feeds, they all have power, fire suppression systems, and they all have ample generator and UPS backup. The two things that we have to do to the physical facilities are, one, clear out dead equipment. There are 22,500 racks of equipment in these facilities that are no longer in service. I'm gonna take you down a walk down memory lane. Titan 5500s from Tellabs, Nortel DMS-250s, Lucent 5ESSs all need to be pulled out. The second thing we need to do is convert these facilities from negative 48 volt DC plant to AC 120. That's just the installation of simple inverters. The capital cost to reposition these networks is fairly de minimis. We are then connecting these networks to our metro footprint.
The Sprint long-distance network of 19,000 route miles of inter-city fiber, 1,100 route miles of metro fiber, terminated literally in the middle of nowhere. Sprint set out to build tandem Class four switch sites, generally 15 or 20 miles from downtown in an industrial area. Its long-distance network, which was built along railroads, terminated in these facilities, and the only two interconnections were with the incumbent carriers, AT&T or Verizon. What we're doing is extending those fiber networks from those facilities to Cogent's 18,000 route mile metro network, and then giving us the ability to sell Wavelengths and transport services in 800 U.S. carrier-neutral facilities. In order to support the transport product, we need to extend the network. In doing so, we add value to the Colo and make it marketable.
Are these colos, would you kind of think of them as, you know, carrier, carrier-neutral, or will you allow other carriers into there?
We absolutely will. Today, Cogent has roughly 630,000 sq ft of raised floor space in 50, six centers that we operate, 55 centers that we operate today. We allow other carriers in. As a practical matter, they don't have the carrier density of, say, an Equinix, but they generally have one or two other players other than Cogent in the facility. We'll do the exact same thing with the Sprint facilities. Then in those facilities, we're not selling value-add services, we're selling raw space, and power, and connectivity. We will sell that space either on a retail basis, rack by rack or on a wholesale basis, data hall by data hall. These are not a brand-new tilt-up million sq ft, you know, facility in Ashburn with 300 megawatts a foot. That's not what these are.
They're also not, you know, a very small cross-connect dense facility, but they're something in between. You know, today they would be called edge computing facilities. They're distributed across the country. They generally have about 120 watts a foot of power, and they're priced inexpensively, along with our network to sell band.
Sorry, what, what's the average power for each facility?
About 120 watts a foot.
If that was fully leased, you know, what, how much revenue you can get per year for, you know, 100 watts, call it?
Today at Cogent, we have roughly 8,000 racks in our 630,000 sq ft facility, and we're generating about $18 million of run rate. If we took the facilities that we're acquiring and had similar occupancy, we'd be looking at an incremental $35 million to $40 million of revenue.
Got it. I mean, it seems like it could be higher than that. I mean, they invested, like, $2 billion in this data center infrastructure.
It could be, and listen, our goal is to maximize value, and we may make this space available either to hyperscalers directly for intermediate-type facilities, or to third parties who will effectively then sell to them.
Just on a wholesale basis.
Got it. Well, yeah, it seems like it's a perfect location to do inferencing at.
Yeah. I mean, if you think about our transaction with Sprint, there were really two very different transactions. The first transaction was we acquired a network for $1 that cost $20 billion to build. You remember when this network was under a contract to be sold to MCI for $129 billion in 2002 and was blocked by the Justice Department. The second thing we acquired is a legacy, large enterprise customer base that was cash flow negative. They were buying a myriad of 28 different products. They were burning almost $1 million a day in servicing those customers. The network itself was basically dormant, empty, but the services that they were selling were very bespoke, kind of managed, customized networks that were very unprofitable.
We're being paid $700 million in cash by T-Mobile to take over that operating business, and we have a handful of objectives: to bring as much traffic on net as we can, to end the life the negative margin products, to modernize the remaining off net services to all fiber delivered, increase port speeds, and maintain revenue. We believe over a four year period, we can take that operating enterprise customer base and convert it to about a 20% EBITDA positive margin business with little or no growth. For any investor who follows legacy enterprise investments, they know that this sector is in decline. Every carrier is experiencing top-line erosion and margin compression.
What, what's the, what's the margin now, right now, roughly the EBITDA margin? Negative 20?
It's actually worse than that. It's more like negative 70.
Okay. Right, that's a big switch.
Well, I mean, one, bringing traffic on net, but the biggest thing is eliminating products that were gross margin negative. One of the deficiencies in that business model was they had no concept of product profitability or customer profitability, so these large enterprise customers effectively had Sprint doing things that were just uneconomical.
Just absolutely just incredible. I mean, what network are they on now, these, these products?
They ride this fiber network, but they generate virtually no traffic. To give you a sense of scale, Cogent today, on its existing network, carries about 1.2, 1.3 exabytes a day of traffic. The entire Sprint customer base was generating less than 10% of that traffic, and most of that came from their few NetCentric customers. Their enterprise customers generated less than one-tenth of 1% of the traffic that Cogent carries. We will just absorb that into our IP backbone.
You're gonna migrate it over to your network. Got it.
Absolutely.
Got it. That makes all the sense in the world. But why, where is the cash going out the door, you know, at this point? Like, why are they losing so much money, this, you know, the 70%, if it's- they have the network up and running, yeah, where does the cash go?
Three big areas. The first one was employees. There used to be over 1,800 employees, at closing, we were down to 950. We will take some additional head count out. two. 93% of their traffic was off-net, so their biggest single expense were very uneconomic off-net loops.
Got it.
We are migrating a lot of that onto the proprietary fiber network that Cogent owns in Metro and reducing those off-net expenses. Then the third major expense was they were running the business with tools and systems that are 20-plus years old. I mean, they were very proud of the fact that they recently just migrated off of mainframes, that many of their systems people still are only proficient in COBOL. You know, these are just kind of models that no longer fit the products and the market we're in. We're migrating those customers into our tool set, migrating those network elements into our network monitoring systems, our trouble ticketing systems, our CRM, and dramatically reducing that back office expense.
Well, maybe going down all the first two, do you have a sense of how many employees you'll end up with, and how long will it take and the timing of that? Also, how long will it take to migrate the traffic onto your own? Can you migrate almost all this onto your own network, and how long will that take you?
Two very different questions. On the employee headcount, when we began negotiating, there were 1,800 employees. By the time we had signed in September of 2022, they were down to 1,320. At closing, we were down to 950. We've already shed about 50 of those 950 acquired, and we have about another 100 to go. In addition to the $700 million we're being paid, T-Mobile is also giving us an additional $25 million for severance. This is really us doing the dirty work, but, you know, not having to pay for it. To the traffic question, they were 93% off-net, 7% on. Cogent, prior to the acquisition, was 75% on-net, 25% off-net.
In analyzing the locations and services, we believe we can bring the Sprint base to about a 50/50 mix, 50% off, 50% off using third-party loops. On those third-party loops, we can become much more efficient by densifying that footprint and exiting some unfavorable take-or-pay agreements. The initial backbone traffic transition has already occurred in the first 90 days, the Sprint legacy network is effectively dormant and ready to sell waves. On the loop work, we're further along internationally. Probably, we're halfway through outside of the U.S., it was a cleaner transition. In the U.S., this is gonna be a several-year project, it's a big reason why we can't improve the margins more quickly. We've said that we'll go from -$300 EBITDA to -$180 in the first 12 months. We'll go from -$80.
$180 to -$80 in the second 12 months. The EBITDA break even after 36 months, and it'll take 48 months to have a 20% EBITDA margin.
Okay, and you don't think there's any way to really accelerate that? It just takes process, it takes time.
It takes time, and they have take-or-pay contracts that we can't exit with loop vendors. That's really the longest pole undetected.
Wow! You didn't really mention, you know, so these enterprise customers, I mean, nothing but nothing, but you can double, triple, quadruple price for some of these services and, you know, their choice at the end of the day, right? I mean, the, you didn't really talk about pricing at all.
In our core transit business, prices will come down perpetually. In the wavelength market, we will be aggressive, but not as aggressive as we have been in transit. We think the superiority of the network, whether it be measured by reach, end of number of endpoints, uniqueness of routes, speed to provision, end-to-end integration, will all be more important than price, but we will use price to gain market share. In terms of the enterprise customers, there are two different things we're going to do. We are going to increase the throughput that they get and standardize them on fiber, eliminating coax, twisted pair, wireless, and satellite. That'll improve the quality, improve the throughput, and keep pricing similar, and there will be certain instances where we will have to raise prices because Sprint had not had the margin discipline that they need.
You know, one thing that's been a little surprising to us since we announced the deal is how quickly the other global internet or enterprise providers have retreated to their home market. Lumen sold off Latin America and Europe. Verizon Global Services announced a large headcount reduction and a draw to its domestic customer base. BT Global Services, similar retrenchment. Orange, similar retrenchment. I do think, to your point, Tim, there's just not a lot of places for these global customers to go, and while it would not be viable to build a global backbone to support these customers, the fact that we have a global transit backbone puts Cogent in a unique position. We operate the largest, most global fiber-based network in the world.
We carry 25% of the world's Internet traffic on that network, but we can easily layer on these enterprise customers at much higher revenue per bit and still fill a need in the market. The final piece of, I think, learning that we've done is how committed many of these customers are to MPLS. We were hoping to convert them to VPLS. We will do some of that conversion, but we were able to look at equipment that we have pulled out of Cogent's [audio distortion], 7, 8-year-old routers that are MPLS-capable. We will be deploying those, we have thousands of them in inventory, to be able to support MPLS for these customers for up to a decade. That message has been very positively received.
It's not, I think, the optimal technology, but for a company that's running a 1,000-endpoint network, the transition costs to a new technology are fairly high, so they're comfortable in staying on that, and we'll support it.
Dave, are you, are you gonna put that legacy Sprint enterprise customer base in, into your existing corporate segment, you know, in terms of, I guess, structure, salespeople, you know, all the operating systems that you have, billing systems, everything else, will, will that get converted in? So you'll end up at the day with really three different units?
Yeah, we are going to use all of the Cogent systems, our CRM system, our billing system, our network management system. Going forward, there will be two very different new categories we'll report on. We'll have a third type of customer. We'll report corporate, enterprise, and NetCentric as three different customer types. We will also add an additional product. Today, we report DIA, VPN, and Colo as products. We will add Wavelengths as a fourth product set. Now, one thing that we discovered after acquiring the asset is the enterprise customer base was not as pure as we were led to believe. There were some NetCentric, some corporate customers in Sprint. Conversely, we looked inside of Cogent and saw that a small number of our corporate customers really fit the definition of enterprise. We are going to reclass those when we report tomorrow.
We will give investors the pro forma of what our growth would have been if we hadn't reclassed, but on a going-forward basis, investors will be able to observe the unit volume and revenue of each of the three customer types and the growth rates of those three different customer base.
Okay, enterprise will be a, a new-
A new line in the press release.
A new line, and then, but you don't, you don't report by product?
We actually do. We usually put it in our investor deck.
Okay
Talk about it on the earnings call. We don't actually put the product breakdown in the press release. We give on net and off net, we give customer type, we give geography, U.S., rest of the world, and we will talk about on the conference call, in every conference call, the breakdown of products.
Great. I mean, why not?
We're adding Wavelengths into the press release because that's such an area of focus of investors.
I mean, personally, Dave, we love more, you know, more of the detail in the press release. Just makes our life a lot easier.
at some point.
We put a lot of detail in. I mean.
I know you do.
You have followed us, and there are two things that are true about Cogent: we try to be transparent, and we try to be consistent. We don't reclass things. We don't change the way we report every couple of years to muddy the waters. We've used the same set of metrics for 18 years. Sprint is forcing us to expand that a bit, but we want to remain consistent. The last point I'll make on that, I won't name specific companies, but there have been companies in our space that actually reported too many metrics, and they did that in order to obfuscate the weakness in their business. I believe strongly in reporting transparently and consistently, report in a salient way so people can really see what's happening in the business.
Well, I, I'm also referring to, you give a lot of detail on the earnings call that's not in the press release. I should kind of clarify a little bit more. We do get a lot, a lot of detail on different locations. It's just nice to have it all in one location.
Yeah, I think you're gonna see the expanded press release will be a little more specific than it has been.
Right.
We've added some additional granularity.
I can't believe you were able to put that all together for this earnings call. You, you, you guys must be working every, every weekend, every night. Congratulations.
Well, the worst thing was we were ready to go last week, but we didn't have an answer from the SEC on reporting until Friday afternoon in August, last Friday. Needless to say, our accounting team was very busy this weekend and getting us ready. We're, we're in a good shape to be able to report tomorrow morning. I look forward to you being on the call. This will be complicated. There are lots of moving pieces, but our commitment is we will be transparent and precise in our answers.
You know, obviously, when you acquire assets, you said there was a few surprises. I'm sure there were some positive and some negative surprises, just kind of like when you get married. Any, you know, on the balance, are things coming out more positive or more negative, or can you say yet? Yeah.
I can say, I think it's on balance about what we expected. I think, you know, the network is better than we expected. The documentation records are better. The facilities had more stuff in them than we expected. The quality of the employees that we kept is better. The quality of the systems and the management was worse than we expected. The customer base was less pure enterprise than we were led to believe. The product mix was probably even worse than we had originally expected in terms of these unviable services. When you kind of push it all together and look in the totality of the transaction, it's about what we expected. I think we're slightly ahead of where we expected to be 90 days in, in part because of T-Mobile.
We were expecting to bring over the IS systems and run them in parallel to Cogent systems. Because of 3 previous data breaches that T-Mobile had had, they've been very reticent in giving us those systems and data, 'cause it's commingled with wireless. They needed to segregate it. It's forced us to put the customers, the network, into the Cogent toolkit much faster than we were thinking. By year-end, we should be entirely operating on a unified platform of Cogent tools.
Well, I mean, that, that might help out on the free cash flow a little bit next year, at least I would think, but not to put words in your mouth, it's kind of a good thing. Is, is T-Mobile going to be, a re they gonna be using you, do you think?
Yeah, they're actually using us two different ways. The first thing is this $700 million contract to buy transit. They are using some of that capacity, but not all of it, by any stretch of the imagination. It's, we've reserved it, it's sitting there ready for them to use as they want. The second thing is, there are a series of other services, mostly Layer two VPN and colocation services, that are outside of that $700 million contract. T-Mobile did become a meaningful customer of Cogent, about $2.5 million-$3 million a month of additional payments for commercial services.
Now, those will go down over time, but I think when you look at where these services are, there's gonna be a very long tail, and I would suspect we'll have a commercial relationship with T-Mobile for more than a decade.
I apologize, that's very helpful. I have a very important question here. We have 1 minute. Have you changed the, the timing of the break even on the EBITDA for the acquired business? I, I think you just said 36 months, if I heard correct. Is it always been 36 months, or?
It has, Tim. We've always said we'll get to, $180 in year to $80 in year two, break even, EBITDA break even, by the end of year three.
I thought that. You know, I absolutely thought that was the case. Someone thought that it was two years. Dave, we're gonna talk tomorrow morning on your earnings, and I'll talk to you afterwards, so you'll be sick of me by the end of the week. I really this is once again just fascinating, a huge amount of information. You know, I kind of encourage you maybe to do an analyst day or something to go through this, 'cause it's unbelievably complicated.
You know, I think we will, Tim. We've actually haven't done one in over a decade. I think we just need to get a little further along in the integration and the reporting, so we can be productive when we sit down with analysts and investors. I do think that's something that we'll look to do maybe in the next six months.
Well, you know, Dave, just my last comment, I mean, you bought those, internet, networks 20 years ago, 21, 22 years ago, you know, right before the internet really took off. You're essentially buying all this AI infrastructure right before AI is really about to take off. You know, best of luck. I know it's a huge risk and a lot, a lot of work, but, you know, it's very, very smart.
Well, thank you very much, Tim. I wanna thank everyone for taking time, and look forward to chatting with you tomorrow, Tim. Take care.
Take care. Bye, everybody.